llm-mcp-server-template

What is llm-mcp-server-template?

llm-mcp-server-template is a template project for developing LLM-MCP (Model Context Protocol) servers. It demonstrates a simple number comparison server and provides example clients that illustrate how various models and invocation methods can interact with MCP servers. It is intended for developers learning to build and integrate MCP tools.

How to use llm-mcp-server-template?

1. Install uv (package manager) and Python 3.10.
2. In the server directory, create a virtual environment and install mcp[cli] and httpx.
3. Use mcp dev math.py to run the server in development mode (supports inspection via mcp-server-inspector).
4. In the client directory, install openai, openai-agents, and socksio, set OPENAI_API_KEY in .env, then run python openai_client.py.

Key features of llm-mcp-server-template

• Implements a number‑comparison MCP server (stdin/stdout transport)
• Includes a working OpenAI client that calls the MCP tool
• Supports inspection with the official mcp-server-inspector
• Provides a clear template for local development and debugging
• Demonstrates the two‑round MCP workflow (tool call + final response)
• Easily extensible to remote transport (SSE) or other hosting platforms

Use cases of llm-mcp-server-template

• Learning how to build a custom MCP server from scratch
• Testing MCP client‑server communication locally before deployment
• Prototyping a new MCP tool that performs simple logic (e.g., compare numbers)
• Integrating an MCP server with AI agents on platforms like OpenAI SDK or LangChain
• Serving as a starter for porting existing command‑line tools to MCP

FAQ from llm-mcp-server-template

What is MCP (Model Context Protocol)?

MCP is a protocol that allows AI models (clients) to call external tools (servers) for capabilities like mathematical computation, file I/O, or network requests. Think of the model as the brain and MCP servers as its tools.

What transport options does llm-mcp-server-template support?

The template currently supports local stdio transport (direct client‑server subprocess). It mentions remote aggregated platforms via SSE (e.g., Glama.ai, MCP.so) as a future direction.

Does the server require Python 3.10?

Yes. The README instructs to install Python 3.10 via uv and to create a virtual environment with that version.

How do I test the server without a client?

You can use mcp-server-inspector by running mcp dev math.py inside the server directory. This provides an interactive UI for exploring tools and calling them.

Can I use this template with Anthropic SDK or LangChain?

The README lists these as future tasks (TODO): using anthropic-sdk, langchain-mcp-adapter, and a fully native implementation. The included client currently uses OpenAI’s openai-agents SDK.